ieee 802.15.4 wpan based intranet chatting

Dynamic Backoff for IEEE 802.15.4 Beaconless Networks

D Rohm, M GoyalUniversity of Wisconsin Milwaukee

Milwaukee, WI 53201 USA{dmgibson,mukul}@uwm.edu

1. Introduction

IEEE 802.15.4 [1] is a leading standard for low powerand low data rate wireless sensor networks. IEEE 802.15.4based wireless technology offers lower installation andmaintenance costs and hence is increasingly replacing theexisting wired technologies in applications such as buildingautomation, home/environment monitoring, industrial con-trol and smart metering. Deployed/future IEEE 802.15.4based networks may range from few nodes in a room toseveral thousand nodes spread sporadically or densely overa large geographical area [3]. The nodes may generate traf-fic infrequently or at a steady rate or in occasional bursts.The overall traffic load on an IEEE 802.15.4 network maybe fairly static or vary unpredictably over a wide range.Clearly, proper configuration is important for successful op-eration of an IEEE 802.15.4 network in given operating con-ditions.

In this paper, we examine the impact of macMinBE/macMaxBE and macMaxCSMABackoffs parameters on theperformance of beaconless operation of IEEE 802.15.4MAC layer under different traffic loads. The performanceis measured in terms of the packet loss probability and thepacket latency. We also develop a dynamic scheme to auto-matically configure the macMinBE, macMaxBE and mac-MaxCSMABackoffs parameters. Our dynamic scheme es-timates the traffic load on the network through examinationof packet loss rates as observed at a particular node. Basedon the perceived network traffic load, maxMinBE, mac-MaxBE, and maxMaxCSMABackoffs are modified. Wepresent data that shows how this dynamic algorithm out-performs the default standard configuration.

The rest of the paper is organized as follows. Section2 provides an overview of the packet transmission processin beaconless IEEE 802.15.4 MAC operation as well as de-scribes different collision scenarios. Section 3 describes thesimulation setup as well as the network performance met-rics used for this study. Sections 4 and 5 present simulationresults analyzing the impact of macMinBE/macMaxBE andmacMaxCSMABackofffs parameters respectively and make

recommendations regarding suitable values for these pa-rameters under different traffic loads. Section 6 presentsa dynamic backoff algorithm and simulation results for thenew algorithm. Section 7 presents a brief survey of previ-ous work on configuring IEEE 802.15.4 networks. Finally,Section 8 concludes the paper.

2. Packet Transmission in Beaconless IEEE802.15.4 MAC Operation: CSMA/CA andRetransmissions

Beaconless IEEE 802.15.4 uses unslotted CSMA/CA. Atransmission attempt begins with a CSMA wait for a ran-dom number of backoff periods between 0 and 2BE − 1,where BE can have a value between macMinBE and mac-MaxBE (by default 3 and 5 respectively). A backoff periodis the time required to transmit 20 symbols, where a symbolis equivalent to 4 bits, on a 250 Kbps channel. Once theCSMA wait is over, the node determines if the channel isavailable for transmission. This clear channel assessment(CCA) is performed over a time duration of 8 symbols. Ifthe CCA fails (i.e. the channel is found to be busy), thenode increments BE (up to macMaxBE), repeats the CSMAwait and the CCA. If the CCA fails even after macMaxC-SMABackoffs (by default 4) re-attempts, a channel accessfailure (CAF) is declared and no further attempt is made tosend the packet. If the CCA succeeds, the node performs anRX-to-TX turnaround1 and transmits the packet.

The packet transmission may get involved in a collision.In the next section, we describe different scenarios that re-sult in a collision in beaconless IEEE 802.15.4 networks.In the absence of a collision, the receiver node receivesthe packet and may optionally send an acknowledgement(ACK) back to the source node. Note that the CSMA/CAprocess is not repeated for the sending of an ACK. The re-ceiving node simply performs an RX-to-TX turnaround of

1The IEEE 802.15.4 nodes are typically half-duplex in nature, i.e. theycan not perform both the transmit (TX) and receive (RX) operations at thesame time. The RX-to-TX or TX-to-RX turnaround time is required to beless than 12 symbols [1].

Copyright 2009 Dawn Rohm and Mukul Goyal

its radio (again up to 12 symbols) and immediately sendsthe ACK. As described in the next section, an ACK mayalso be involved in a collision and thus get lost. The resultof an ACK collision is the same as that of a packet collision.If an ACK is required, the source node reattempts to sendthe packet after waiting for macAckWaitDuration symbols(54 symbols for 2.4 GHz PHY operation) after finishing thepacket transmission. A failure is declared if no ACK is re-ceived even after macMaxFrameRetries (by default 3) re-transmissions. Such a failure is referred to as the collisionfailure in the subsequent discussion.

In beaconless IEEE 802.15.4 networks, collisions maytake place either due to hidden nodes or due to non-negligible RX-to-TX (and TX-to-RX) turnaround times.

Hidden nodes: Some nodes in the network may not bein the hearing range of a node (say node X) and hence maytransmit a packet at the same time as node X. Such nodes arecalled hidden nodes for node X. If node Y, the destination ofnode X’s transmissions, can hear these hidden nodes, anyconcurrent transmission by a hidden node would cause nodeY to drop node X’s transmission.

Collisions due to turnaround time: As mentioned earlier,an IEEE 802.15.4 node may take up to 12 symbols to turnaround from RX mode to TX mode and vice-versa. Thisnon-negligible turnaround time may cause packet collisionsto take place in the following situations:

1) Suppose, a number of nodes, all in each other’s hear-ing range, are competing for channel access and all of themare doing the CSMA wait at a certain time, hence the trans-mission channel is idle. Suppose, node A is the first nodeto wake up at time t. Node A performs a CCA till timet + 8, which is guaranteed to succeed, and then performsan RX-to-TX turnaround that finishes at time t + 20. Thetransmission channel would continue to be idle until timet + 20 when node A begins its packet transmission. Thus,if another node finishes its CSMA wait between times t andt + 12, its CCA would succeed and its subsequent packettransmission would collide with that of node A. Figure 1(a)refers to this 12 symbol duration as the first collision win-dow. Note that the first collision window is actually equalto the RX-to-TX turnaround time.

2) A destination node (say B) needs to complete an RX-to-TX turnaround before it can send the acknowledgementfor a packet. If another node finishes its CSMA wait duringthe first 4 symbols of this turnaround, its CCA would suc-ceed and its packet transmission would collide with nodeB’s acknowledgement. Figure 1(b) refers to this 4 sym-bol duration as the second collision window. Clearly, thesecond collision window is the result of CCA duration be-ing less than the RX-to-TX turnaround time.To eliminatethe second congestion window, we suggest increasing theCCA duration to a value larger than 12 symbol RX-to-TXturnaround time. The same suggestion has been indepen-

8 Symbols 12 symbols

Node A finishes CSMA wait Node A finishes turnaround

Node A finishes CCA

(12 symbols)First Collision Window

(a) First Collision Window

12 symbols

pkt transmissionNode A finishes

receive the ACKNode A begins to

(4 symbols)Second Collision Window

(b) Second Collision Window

Figure 1. Collision Windows

dently made in [5]. Note that the second collision windowexists only if no collision takes place in the first collisionwindow.

3) A destination node would ignore a packet transmis-sion if it begins before the destination has completed theTX-to-RX turnaround after sending the acknowledgementfor the previous transmission. Even though this situationdoes not involve a collision, its impact is same as that of acollision.

3. Simulation Setup and Performance Metrics

The simulations make the following assumptions. TheIEEE 802.15.4 MAC layer operates in the beaconless modeand all the packets require MAC level acknowledgement.The CCA is performed over 16 symbols to ensure that anACK is never involved in a collision. The IEEE 802.15.4PHY layer operates in 2.4GHz band and no transmissionis lost due to signal attenuation/corruption over the chan-nel. In this paper, we investigate the scenario where hiddennodes are not a significant problem. Hence, in our simula-tions, we ensure that all the nodes are in each other’s radiorange and thus there are no hidden nodes.2 Each node gen-erates traffic for the common coordinator as per a poissondistribution with average rate 5 packets/second. The sim-ulations were performed with several different packet sizesalthough the results presented here were obtained using 133byte long packets, which is the maximum allowed size foran IEEE 802.15.4 PHY frame including the 5 byte synchro-nization header and 1 byte PHY header [1].

The traffic load across simulations is varied by chang-ing the number of nodes, excluding the coordinator, in therange {10, 12, . . . , 22, 26, . . . , 38, 40, 50, 60}. In each sim-ulation, a certain number of nodes (between 10 and 60)

2Understanding how to counter the impact of hidden nodes by properlyconfiguring IEEE 802.15.4 parameters is a part of our ongoing research.

2


send packets to their common coordinator. Since each nodegenerates on average 5 packets/second, we simulate aver-age traffic loads ranging from 50 to 300 packets/second.Note that with 133 byte packet size, a packet transmis-sion takes channel time of 300 symbols (266 symbols forpacket transmission + 12 symbols for receiver’s RX-to-TXturnaround + 22 symbols for 11 byte acknowledgementtransmission). Hence a 2.4 GHz channel, with channel ca-pacity 250 Kbps (or 62500 symbols/second), can carry atmost 208.33 (= 62500/300) packets per second. Thus,the simulations cover a wide range of traffic load scenariosfrom a lightly loaded network (50 packets/second) to sig-nificantly overloaded network (300 packets/second). Addi-tionally, performance data was only collected after the net-work reached a steady state where association with the coor-dinator and route discoveries have already been completed.

In order to analyze how the dynamic algorithm reactsto changes in the network traffic load, we also performedsimulations with variable traffic loads. Our variable ratesimulations also wait until after the network has reached asteady state before collecting data. Variable rate simulationsstart with a traffic load of 50 pps which is maintained for100 seconds. Every 100 seconds the traffic load changes.The traffic loads through the simulation are 50, 100, 150,300, 150, 100, 300, 100 packets per second starting at times9100, 9200, 9300, 9400, 9500, 9600, 9700, and 9800 re-spectively.

The network performance metrics used in this study arethe packet loss probability and the packet latency. Thepacket loss probability is the probability that the MAC layerfails to send a packet to its destination. As discussed earlier,the packet loss can take place due to a channel access fail-ure (CAF) or a collision failure. The packet loss probabilityfor a node is simply the fraction of packets lost by the nodeduring the simulation run. The CAF probability is the prob-ability that a packet encounters (1 + macMaxCSMABack-offs) consecutive CCA failures. The CAF probability fora node is calculated as the number of CAFs it suffers di-vided by the total number of transmission attempts it makes.The probability of collision for a transmission by a node iscalculated as the ratio of total number of collisions experi-enced by the node during the simulation and the total num-ber of transmissions (transmission attempts excluding theones that ended in CAF) it makes. Note that the probabilityof collision for a transmission is not the same as the prob-ability of collision failure. A collision failure occurs onlywhen (1 + macMaxFrameRetries) back-to-back collisionstake place during the transmission of a packet or its ACK.The packet latency is defined as the time interval betweenthe instants when the IEEE 802.15.4 MAC layer receivesa packet for transmission and when it reports the successor failure in sending the packet back to the higher layer.The packet latency for a node is calculated as the average

latency for the packet it generates. The performance met-rics values reported in the subsequent sections are averagesacross all the nodes in the simulation. The 95% confidenceintervals associated with these values were always observedto be within a few percentage of the average.

As described in the previous section, in IEEE 802.15.4MAC operation, the CSMA wait time depends on the BEvalue. For each transmission attempt, BE is initialized tomacMinBE and each CCA failure causes it to increase by 1until it reaches a maximum (macMaxBE). Thus, the CSMAwait duration depends on how many CCA failures have al-ready taken place in the current transmission attempt. Tosimplify our investigation, we eliminate this dependencyby setting macMinBE and macMaxBE to the same value,referred to henceforth simply as BE. Thus, in our simula-tions, the CSMA wait time simply depend on BE irrespec-tive of the CCA failures experienced so far in the currenttransmission attempt. In this study, we experimented withmacMinBE(=macMaxBE) values 3 through 8. The value ofthe macMaxCSMABackoffs parameter used in the simula-tions varied in range 0 through 73. Although we performedsimulations with different values of the macMaxFrameRe-tries parameter as well, the parameter was observed not tohave much impact on the performance. We believe that thetrue impact of the macMaxFrameRetries parameter can onlybe determined when collisions are a frequent occurance inthe network, i.e. in the presence of a significant number ofhidden nodes. Analyzing the impact of macMaxFrameRe-tries parameter on beaconless IEEE 802.15.4 performancein the presence of hidden nodes is part of our ongoing re-search.

4. Impact of BE (macMinBE/macMaxBE)Value

Figure 2 shows the impact of increasing the BE valueon different performance metrics as the trafic load onthe network increases. In these simulations, the mac-MaxCSMABackoffs and macMaxFrameRetries parametersare maintained at their default values (4 and 3 respectively).Figure 2(a) reveals that, at low traffic loads, the increase inBE can significantly reduce the packet loss probability fora given traffic load. However, as the traffic load increases,the reduction in the packet loss probability with increasein BE becomes less significant. At very high traffic loads,the packet loss probability becomes very high irrespectiveof the BE value. The CAF probability for a transmissionfollows essentially the same trend as the packet loss proba-bility (Figure 2(b)).

3Although IEEE 802.15.4 specification [1] limits macMaxCSMABack-offs to a maximum value of 5, we found merit in increasing the parameter’svalue beyond this limit (Section 5).

3


0

0.1

0.2

0.3

0.4

0.5

0.6

300 200 100 50

Pac

ket L

oss

Pro

babi

lity

Traffic Load (pkts/s)

BE=3BE=4BE=5BE=6BE=7BE=8

(a)

0 0.05

0.1 0.15

0.2 0.25

0.3 0.35

0.4 0.45

300 200 100 50

CA

F P

roba

bilit

y



(b)

0 0.05

0.1 0.15

0.2 0.25

0.3 0.35

0.4 0.45

0.5

300 200 100 50

Col

lisio

n P

roba

bilit

y



(c)

0 20 40 60 80

100 120 140 160 180 200 220

300 200 100 50

Pac

ket L

aten

cy (

ms)



(d)

Figure 2. Impact of macMinBE/macMaxBEvalue

These observations can be explained as follows. Sup-pose certain nodes are competing for channel access at acertain time. At low traffic loads, the size of this set is smalland there are no new additions to it, i.e. no new node getsa packet to send, for a relatively long time. The increase inBE increases the range of CSMA wait times, which in turncauses the packet transmissions to be spread throughout thistime. Thus, a node becomes less likely to sense the trans-mission channel while another node is in middle of a trans-mission. Moreover, as packets are successfully transmitted,there is less competition for channel access and hence theCAF probability goes down. At high traffic loads, the num-ber of nodes competing for channel access at a certain timemay be large and new nodes continuously enter the set ofcompeting nodes as some nodes leave. Thus, increasing therange of CSMA wait times to spread out the packet trans-missions does not help much.

Similar to the CAF probability, the collision probabilityalso reduces with increase in BE at low traffic loads, al-though the reduction is not as significant as for the CAFprobability (Figure 2(c)). At high traffic loads, the colli-sion probability increases slightly with increase in BE. Aswe discussed earlier, in the absence of hidden nodes, thenon-negligible RX-to-TX turnaround time is the reason col-lisions take place. At low traffic loads, increase in BE in-creases the range of CSMA wait times. Thus, a node isless likely to finish its CSMA wait during the turnaroundtime of a node about to begin its packet or ACK transmis-sion. Since the turnaround time is required to be less than12 symbols [1], the increase in range of CSMA wait timescauses only a modest decrease in the collision probability.The slight increase in the collision probability with increasein BE at high traffic loads can be due to several factors.First, higher BE values result in slightly lower CAF prob-ability even at high traffic loads. Thus, higher BE valuescause more packet transmissions which may result in morecollisions. Secondly, higher BE values increase the packetlatency which means that the number of nodes competingfor channel access at any given time increases, which againresults in more collisions.

Figure 2(d) shows that, despite lower collision rates atlow traffic loads, the packet latency is consistently higherfor higher BE values. This is because the longer CSMAwaits overshadow the effect of fewer retransmissions due tocollisions. As mentioned earlier, BE values 7 and 8 resultin lowest packet losses but produce such high latency valuesthat they are likely to be useful only for those applicationsthat can tolerate very high latencies. However, as Figure2(a) shows, BE value 6 does not significantly increase thepacket loss probability compared to a BE value of 7 or 8.For this reason, BE value 6 is likely to be best for the ap-plications that need low packet loss rates and can tolerateper-hop packet latencies up to 50 ms. Applications with

4


more stringent latency requirements may benefit from BEvalue 5. Lower BE values would be useful for only thoseapplications that can tolerate packet losses but need mini-mum packet latencies.

The results shown in Figure 2 were obtained using 133byte long packets. The simulations were repeated with sev-eral smaller packet sizes as well and the results obtainedwere qualitatively similar. In particular, we always foundBE values 5 and 6 to offer the best tradeoff between thepacket loss probability and the latency.

5. Impact of macMaxCSMABackoffs Value

The simulation results regarding the impact of mac-MaxCSMABackoffs value on beaconless IEEE 802.15.4 op-eration are displayed in Figure 3. The label backoffs in thefigures refers to macMaxCSMABackoffs. In these simula-tions, the macMinBE and macMaxBE parameters are set tovalue 5 each and macMaxFrameRetries parameter is set toits default value 3. To allow easy observation of main re-sults, we show curves for macMaxCSMABackoffs values 2,4 and 7 only.

It is clear that the increase in the macMaxCSMABackoffsvalue reduces the CAF probability across all traffic loads(Figure 3(b)) as more CCA failures are allowed in a trans-mission attempt before a channel access failure is declared.Reduction in the CAF probability means that more trans-missions take place, which translates to a higher probabilityof collision for a transmission (Figure 3(c)). However, asFigure 3(c) shows, the increase in the probability of colli-sion, with increase in macMaxCSMABackoffs value, is notsubstantial for traffic loads up to 100 packets/sec. The de-crease in the CAF probability dominates the increase in thecollision probability and causes the overall packet loss prob-ability to go down with increase in macMaxCSMABackoffsvalue for traffic loads up to 200 packets/sec. For highertraffic loads, the increase in the collision probability be-comes large enough to neutralize the impact of reducedCAF probability and the overall packet loss probability be-comes slightly higher for larger macMaxCSMABackoffs val-ues.

The increase in the macMaxCSMABackoffs value alsocauses an increase in the packet latency (Figure 3(d)), whichbecomes significant at higher traffic loads, as a packet hasless chance to be abandoned because of a channel accessfailure and more chance to be retransmitted due to colli-sions. The substantial increase in the packet latency and thecollision probability at higher traffic loads, with increase inthe macMaxCSMABackoffs value, can be attributed to theirmutual dependence. Higher packet latency means that apacket competes with a larger number of other packets foraccess to the transmission channel, which results in morecollisions. More collisions, in turn, mean more retransmis-

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

300 200 100 50

Pac

ket L

oss

Pro

babi

lity


backoffs=2backoffs=4backoffs=7

(a)

0

0.1

0.2

0.3

0.4

0.5

0.6

50 100 200 300C

AF

Pro

babi

lity



(b)

0

0.1

0.2

0.3

0.4

0.5

0.6

300 200 100 50

Col

lisio

n P

roba

bilit

y



(c)

10

15

20

25

30

35

40

45

50

300 200 100 50

Pac

ket L

aten

cy (

ms)



(d)

Figure 3. Impact of macMaxCSMABackoffs valuewith Channel Access Failures allowed

5


0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

300 200 100 50

Pac

ket L

oss

Pro

babi

lity


noCAF,backoffs=2noCAF,backoffs=4noCAF,backoffs=7

(a)

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

60 80 100 120 140

Pac

ket L

oss

Pro

babi

lity



backoffs=7

(b) CAF allowed versus CAF disabled

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

1

300 200 100 50

Col

lisio

n P

roba

bilit

y



(c)

0

20

40

60

80

100

120

140

300 200 100 50

Pac

ket L

aten

cy (

ms)



(d)

Figure 4. Impact of macMaxCSMABackoffs valuewith Channel Access Failures disabled

sions of a packet and hence higher latency. Simulations withsmaller packet sizes revealed essentially the same trends asdescribed above with the differences easily accounted forby the change in the packet size.

The simulation results show that setting the macMaxC-SMABackoffs parameter to value 7 results in a reasonablepacket latency (less than 30ms) and lower packet loss ratesthan other experimented values for traffic loads up to a cer-tain threshold. This threshold value is observed to be about150 packets/sec for 133 byte long packets and gets higherfor smaller packet sizes. As IEEE 802.15.4 specification[1] limits the macMaxCSMABackoffs parameter to a maxi-mum value 5, we suggest modifying the standard to allowhigher values for the parameter. At traffic loads higher thanthis threshold, setting macMaxCSMABackoffs parameter tovalue 4, which is also the default value for the parameter,gives the best tradeoff between the packet loss rate and thepacket latency.

The results discussed above suggest that, at low andmoderate traffic loads, further reductions in the packet lossprobability can be obtained if we eliminate the restrictionimposed by the IEEE 802.15.4 standard on how many timesa node can perform CCA without success while attemptingto send a packet. The IEEE 802.15.4 standard requires apacket to be abdandoned if (1 + macMaxCSMABackoffs)consecutive CCA failures take place, characterizing the sit-uation as a channel access failure. Rather than abandoningthe packet, we consider treating a channel access failure thesame way as a collision. Under this modification, a nodewould start the next attempt to send the packet (resetting theNB parameter to 0 and BE to macMinBE) when it encoun-ters (1 + macMaxCSMABackoffs) consecutive CCA failuresin the current transmission attempt or when it fails to re-ceive an acknowledgement for the packet transmission. Asbefore, a node can make atmost (1 + macMaxFrameRetries)attempts to send a packet.

The simulation results under the modified scheme areshown in Figure 4. The performance curves for the mod-ified scheme are labeled with prefix noCAF. As Figure 4(b)shows, the modification indeed leads to lower loss prob-ability than the standard scheme for traffic loads up to athreshold (140 packets/second for 133 byte packets). Notethat the packet loss probability values under the modifiedscheme with macMaxCSMABackoffs value 2 are very closeto the values under the standard scheme with macMaxCS-MABackoffs value 7. Increasing the macMaxCSMABack-offs value from 2 to 4 and then to 7 under the modifiedscheme causes continuous, albeit diminishing, reductionsin the packet loss probability for traffic loads up to the 140packets/second threshold. At higher traffic loads, the packetloss probability increases with increase in the macMaxCS-MABackoffs value (Figure 4(a)).

The observed change in the packet loss probability for

6


Figure 5. Dynamic Algorithm

different macMaxCSMABackoffs values under the modi-fied scheme can be explained on the basis of the observedchange in the collision probability. Figure 4(c) displaysthe collision probability for a transmission for differentmacMaxCSMABackoffs values under the modified scheme,where the erstwhile channel access failures are also countedas the collisions. Note that, for traffic loads up to the140 packets/second threshold, the collision probability de-creases with increases in macMaxCSMABackoffs value andvice versa for higher traffic loads. Clearly, for traffic loadsup to the threshold, the decrease in the number of erstwhilechannel access failures with higher macMaxCSMABackoffsvalues dominates any increase in the actual collisions dueto more transmissions taking place. At higher traffic loads,the impact of increase in the number of actual collisionsdominates. Note that, under the modified scheme, the in-crease in the macMaxCSMABackoffs value leads to signif-icant increase in the packet latency at traffic loads higherthan the 140 packets/second threshold (Figure 4(d)). Simu-lations with smaller packet sizes gave similar results exceptthat the threshold traffic load was larger.

A comparison of Figure 3(a) and Figure 4(a) revealsthat, for 133 byte long packets, the modified scheme re-sults in lower loss rates than the standard scheme for traf-fic loads up to a threshold (about 150 packets/second) andvice versa for the higher traffic loads. Further, an examina-tion of Figures 4(b) and 4(d) reveals that under the modifiedscheme, setting macMaxCSMABackoffs to value 4 gives thebest tradeoff between the packet loss probability and thepacket latency for traffic loads up to the threshold. Simi-lar behavior was observed for lower packet sizes with thethreshold traffic load being correspondingly higher. Sincean IEEE 802.15.4 network can be expected to operate in thelow/moderate (less than the threshold) traffic load regimemost of the time, there is some merit in modifying the IEEE802.15.4 standard to include the modified scheme as a con-figurable option.

6. Dynamic Backoff Algorithm

The proposed dynamic backoff algorithm tunes the pa-rameters macMinBE, macMaxBE, and macMaxCSMABack-offs for currently observed network traffic loads. The goalof the algorithm is to decrease packet loss rates and in-crease throughput while maintaining reasonable latencies.To this end, we attempt to maintain average latency valuesof 40 ms or below. Note that this algorithm could easilybe adapted to an alternative acceptable latency value. Ouralgorithm estimates the current network traffic load by mon-itoring packet loss rates over previous packets. Centralizedalgorithms which may be able to calculate such informationprecisely involve more overhead than is practical for IEEE802.15.4 devices where energy conservation is critical. As aresult, the proposed algorithm estimates the recent networkby observing the results of previous transmission attempts.

The state diagram depicting our dynamic algorithm canbe seen in figure 5. For networks with very low traffic rates(50 to 140 packet/s in our simulations), the dynamic algo-rithm uses a configuration of macMinBE/macMaxBE=6 andmacMaxCSMABackoffs=7. This configuration is refered toas state 1 in figure 5. At approximately 140 packets/s, theaverage latency begins exceeding the 40 ms threshold. Atthis point, the algorithm detects the higher latency valuesand transitions to state 2 (macMinBE/macMaxBE=5, mac-MaxCSMABackoffs=7) in an effort to lower the packet la-tencies. It can be noted that in terms of packet loss, state 1remains optimal above the 140 packets per second thresh-old, but because the latency values become excessive, itis necessary to transition to state 2. Additionally, if aver-age packet loss rates jump above 30%, indicating a trafficload exceeding 200 packets/s, transition is made to state 3(macMinBE/macMaxBE=5, macMaxCSMABackoffs=4).

From state 2, transition can be made back to state 1 ifboth the packet loss rate drops below 15% and the latencybelow 40 ms. A packet loss rate below 15% indicates thatthe traffic load has dropped below 140 packets per second,and thus the state 1 configuration is again optimal. The re-quirement that the average packet latency be below 40 msis a safe guard to ensure the latency requirement and to pre-vent a possibly ping pong effect between state 1 and statefor a network with a constant traffic load. State 2 will tran-sition to state 3 if either the loss rate reaches 30%, indicat-ing a traffic load exceeding 200 packets/s, or if the averagelatency still exceeds 40 ms. Finally, from state 3, the algo-rithm will transition to state 1 if the traffic load goes below140 packets/s (loss rate below 15%) and the average latencyis below 40 ms.

State 3 transitions to state 2 if the average latency is be-low 40 ms and the traffic load is less than 200 packets/s(30% packet loss) but greater than 140 packets/s (packetloss of 15%). State 3 may otherwise transition to state 1

7


0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

50 100 150 200 250 300

Pac

ket L

oss

Pro

babi

lity


Default,CCA=8Default,CCA=16

Dynamic

(a) Packet Loss

5

10

15

20

25

30

35

50 100 150 200 250 300

Late

ncy

(ms)


Default,CCA=8Default,CCA=16

Dynamic

(b) Latency

Figure 6. Dynamic vs. Standard

if the traffic load drops below 140 packets/s (15% packetloss) and the average latency is below 40 ms.

Figure 6(a) shows much improved performance forour Dynamic algorithm over the standard default set-tings. The improvement is significant even when youcompare our dynamic algorithm to the standard defaultconfiguration with a clear channel assessment (CCA)time of 16 instead of the standard 8. This shows that theimprovement is not just due to increasing the CCA timeto 16, but a result of our dynamic configuration. Thedynamic algorithm does result in higher latencies thanthe IEEE 802.15.4 standard implementation; however,the decrease in packet loss and increase in throughputoutweighs the disadvantage of increased latencies formost applications, and again, the algorithm can be easilyreconfigured to tolerate higher or lower latency values. Thelargest improvement in packet loss rates is seen at lowertraffic rates. This is a result of the increase of backoffexponent values to 6 and the number of CSMA backoffsto 7. At higher traffic rates, the results of our dynamicalgorithm are similar to the standard using a CCA value of16. This is because the configuration at high traffic loads(macMinBE/macMaxBE=5,macMaxCSMABackoffs=4)

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

50 100150

300150

100300

100

Pac

ket L

oss

Pro

babi

lity

(ms)


Standard(CCA=8)Standard(CCA=16)

Dynamic

(a) Packet Loss

5

10

15

20

25

30

35

40

50 100150

300150

100300

100P

acke

t Lat

ency

(m

s)


Standard(CCA=8)Standard(CCA=16)

Dynamic

(b) Latency

Figure 7. Variable Traffic Load Performance

is a near match to the standard configuration(macMinBE=3,macMaxBE=5,macMaxCSMABackoffs=4).

In variable rate simulations, our dynamic algorithmquickly adapts to changes in network traffic levels. Figures7(a) and 7(b) show average values for all nodes over the100 second time period that a given traffic load is in effect.As can be seen in figure 7(a), our dynamic algorithm re-acts and outperforms the standard default in most cases. Atlower traffic load levels, our algorithm again shows signifi-cant improvement over the standard. At higher traffic loads,our algorithm performs similarly compared to the standard.7(b) shows average latency values over these 100 secondtimeframes staying below 35 ms for all traffic loads. This iseven better than our requirement of staying below a 40 msaverage.

As can be seen in figure 8(a), and 8(b) this dynamic algo-rithm does a good job of estimating the current network traf-fic load and adjusting its configuration. 8(a) shows nearlyperfect matches to the component configurations. Latencyvalues match exactly at lower and higher traffic rates, whilethe midrange traffic loads gradually follow the trend as ittransitions. This is because the final dynamic algorithm weimplemented moves to state 2 at an earlier point in an effort

8


0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

50 100 150 200 250 300

Pac

ket L

oss

Pro

babi

lity


BE=5,Backoffs=7BE=6,Backoffs=7Default,CCA=16

Dynamic

(a) Packet Loss

0

20

40

60

80

100

120

50 100 150 200 250 300

Pac

ket L

aten

cy (

ms)


BE=5,Backoffs=7BE=6,Backoffs=7Default,CCA=16

Dynamic

(b) Latency

Figure 8. Dynamic Algorithm

to maintain average latency values below 40 ms.The configuration window of 40 packets was carefully

chosen to allow the dynamic algorithm to react quicklyto changes in the network traffic load while also avoid-ing bouncing between configuration states. The smallerthe window of packets between configuration changes, themore responsive the algorithm is to changes in the net-work. For example, with the jump from 100 packet/s to300 packet/s, the response times were 2, 5 and 13 secondsfor window sizes of 20, 40, and 80 respectively. The changefrom 300 packet/s to 100 packet/s takes even longer to ad-just the configuration with 21, 28, and 67 seconds respec-tively. The response time when analyzing 80 packets at atime is likely too slow for most applications. While the re-sponse time for a window size of 20 packets is faster, figure9(a) shows the algorithm jumping around as it reacts toohastily. For these reasons, we believe that a window size of40 packets will provide the best performance.

7. Related Work

The use of IEEE 802.15.4 standard in commercial ap-plications is still at an early stage and hence there arenot many papers that investigate proper configuration ofIEEE 802.15.4 MAC paremeters. Koubaa et.al.[8] ob-

1

2

3

Alg

orith

m S

tate

Time

(a) Window=20 packets

1

2

3

Alg

orith

m S

tate

Time

(b) Window=40 packets

1

2

3

Alg

orith

m S

tate

Time

(c) Window=80 packets

Figure 9. Algorithm States for a Single Node

served that the packet loss rate in a beacon enabled net-work can be reduced at the cost of increasing the packetlatency by increasing the macMinBE/macMaxBE values.Tao et. al. [13] observed that, under saturated4 condi-tions in beacon enabled networks of more than 4 nodes, in-crease in macMinBE/macMaxBE values from (3/5, 3/3) to(4/6, 5/5) respectively improves the network throughput.

Additionally, many papers prescribe using different BEvalues to achieve service differentiation. Koubaa et. al.[7]suggest using lower macMinBE values to achieve low la-tency for time critical traffic in beacon mode IEEE 802.15.4networks. Ko et.al. [6] suggest allowing nodes that needto transmit frequently to use lower than normal macMinBE.Youn et.al. [14] suggest achieving priority based servicedifferentiation in IEEE 802.15.4 networks by chosing theCSMA wait duration using different gaussian distributionfor different priorities. Ha et.al. [4] suggest a scheme fordetermining BE value for a new send attempt based on thefinal BE value reached in the previous send attempt. Finally,there are many papers [10, 9, 11, 12] that require the coor-dinator to dynamically assign BE values to the associateddevices since the coordinator may have access to useful in-formation such as the individual/total traffic loads and thenumber of nodes in the cluster.

As per our literature search, only a few papers have sofar analyzed the impact of macMaxCSMABackoffs value onIEEE 802.15.4 operation. Athanasopoulos et.al. [2] sug-gested using macMaxCSMABackoffs value 1, rather thandefault 4, to reduce the power consumption and the packet

4where a node always has a packet to send

9


latency while ignoring any detrimental effect such a settingmay have on the packet loss rates. Tao et.al. [13] ob-served that macMinBE/macMaxBE parameters have moredirect influence on the network throughput than macMaxC-SMABackoffs parameter.

8. Conclusion

In this paper, we analyzed the impact ofmacMinBE/macMaxBE and macMaxCSMABackoffs pa-rameters on the performance of beaconless IEEE 802.15.4networks. We have also proposed a dynamic algorithm thatadjusts itself to observed changes in the network trafficload. Network traffic loads are estimated by monitoringlatency and packet loss rates at each individual node. Thisis important as it avoids additional communications forthe nodes that would be necessary in a centrally managedalgorithm. Finally, we have shown that this algorithm isable to quickly adapt to changes in the observed networktraffic load.

References

[1] IEEE standard for information technology- telecommuni-cations and information exchange between systems- localand metropolitan area networks- specific requirements part15.4: Wireless medium access control (MAC) and physi-cal layer (PHY) specifications for low-rate wireless personalarea networks (WPANs). IEEE Std 802.15.4-2006 (Revisionof IEEE Std 802.15.4-2003), 2006.

[2] A. Athanasopoulos, E. Topalis, C. Antonopoulos, andS. Koubias. 802.15.4: The effect of different back-offschemes on power and qos characteristics. Wireless andMobile Communications, 2007. ICWMC ’07. Third Interna-tional Conference on, pages 68–68, March 2007.

[3] M. Dohler and T. Watteyne. Urban WSNs Routing Re-quirements in Low Power and Lossy Networks. Internet-Draft draft-ietf-roll-urban-routing-reqs-00, Internet Engi-neering Task Force, Sept. 2008. Work in progress.

[4] J. Y. Ha, T. Kim, H. S. Park, S. Choi, and W. H. Kwon.An enhanced CSMA-CA algorithm for IEEE 802.15.4 LR-WPANs. Communications Letters, IEEE, 11(5):461–463,May 2007.

[5] T. O. Kim, J. S. Park, H. J. Chong, K. J. Kim, and B. D.Choi. Performance analysis of IEEE 802.15.4 non-beaconmode with the unslotted CSMA/CA. IEEE CommunicationsLetters, 12(4):238–240, Apr. 2008.

[6] J.-G. Ko, Y.-H. Cho, and H. Kim. Performance evalua-tion of IEEE 802.15.4 MAC with different backoff rangesin wireless sensor networks. Communication systems, 2006.ICCS 2006. 10th IEEE Singapore International Conferenceon, pages 1–5, Oct. 2006.

[7] A. Koubaa, M. Alves, B. Nefzi, and Y. Song. Improvingthe IEEE 802.15.4 slotted CSMA/CA MAC for time-criticalevents in wireless sensor networks. Workshop of Real-TimeNetworks, Satellite Workshop to ECRTS 2006, July 2006.

[8] A. Koubaa, M. Alves, and E. Tovar. A comprehensivesimulation study of slotted CSMA/CA for IEEE 802.15.4wireless sensor networks. Factory Communication Systems,2006 IEEE International Workshop on, pages 183–192, 27,2006.

[9] B.-H. Lee and H.-K. Wu. A delayed backoff algorithm forIEEE 802.15.4 beacon-enabled LR-WPAN. Information,Communications and Signal Processing, 2007 6th Interna-tional Conference on, pages 1–4, Dec. 2007.

[10] A.-C. Pang and H.-W. Tseng. Dynamic backoff for wirelesspersonal networks. Global Telecommunications Conference,2004. GLOBECOM ’04. IEEE, 3:1580–1584 Vol.3, Nov.-3Dec. 2004.

[11] H. M. Park, W. C. Park, S. J. Lee, and G.-Y. Lee. Modi-fied backoff scheme for MAC performance enhancement inIEEE 802.15.4 sensor network. In 2007 WSEAS Interna-tional Conference on CIRCUITS, SYSTEMS, SIGNAL andTELECOMMUNICATIONS, January 2007.

[12] V. P. Rao and D. Marandin. Adaptive backoff exponent al-gorithm for Zigbee (IEEE 802.15.4). In Next GenerationTeletraffic and Wired/Wireless Advanced Networking, vol-ume 4003 of Lecture Notes in Computer Science, pages 501–516. Springer-Verlag, 2006.

[13] Z. Tao, S. Panwar, D. Gu, and J. Zhang. Performance anal-ysis and a proposed improvement for the IEEE 802.15.4contention access period. Wireless Communications andNetworking Conference, 2006. WCNC 2006. IEEE, 4:1811–1818, 0-0 2006.

[14] M. Youn, Y.-Y. Oh, J. Lee, and Y. Kim. IEEE 802.15.4 basedQoS support slotted CSMA/CA MAC for wireless sensornetworks. Sensor Technologies and Applications, 2007. Sen-sorComm 2007. International Conference on, pages 113–117, Oct. 2007.

10


ieee 802.15.4 wpan based intranet chatting

Documents